Skip to content

Cards

When a result is worth seeing, the agent shows it as a Card in its answer, right after it queried a Data Source. A Card has four tabs:

  • Chart: the result drawn as a chart. Point at a bar, line or slice to see its values.
  • Table: the rows of the result, at most 10,000. A wide table scrolls sideways inside the Card.
  • Query: which Data Source the result came from, and the SQL the agent ran. For a Card that didn’t come from a query, this tab is called Source.
  • Spec: the chart’s Vega-Lite spec, with the changes you made. This is what the agent sees of the chart.

The Card opens on the tab the agent chose: the Chart for most results, the Table for a list that reads better as rows. A dot marks that tab.

You can look at a result another way without asking the agent again:

  • Chart type: where the agent offered more than one chart, such as bars and a pie, pick one above the chart.
  • Options: where the agent added options, such as “Top N” or “Granularity”, change them above the chart. The chart follows at once.
  • Tooltips, zoom and series: point at a bar, line or slice to see its values. Where the agent allowed it, scroll to zoom and drag to pan. Click an entry of the legend to show only that series, and click it again to show all. Until you first use a chart, a hint says what it can do.
  • Table: click a column’s header to sort by it; click again to reverse, and a third time to undo. Type into the field below a header to filter that column. For numbers you can also compare, such as > 1000 or <= 50. The Card says how many rows are left.
  • Settings: change the colours, where the legend sits or whether it shows (offered when the chart has a legend), and the titles of the axes. The chart shows each change as you make it. Settings change only how the chart looks; to change what it shows, use the options or ask the agent. Reset goes back to the agent’s chart.

Your changes are saved with the Card: they are still there after a reload, and the agent sees the chart as you left it in your next message.

  • Copy PNG copies the chart to the clipboard as a picture, light on white, ready to paste into a slide or a message.
  • Excel downloads the table as an Excel file, with the filters you set. Its name is the Card’s title and the date and time, such as Revenue by region 2026-11-10 09-05-07.xlsx.
  • Fullscreen opens the Card over the whole window, for a dense chart or a wide table. Press Escape to close it.

If a Card is wrong or isn’t what you wanted, point the agent at exactly that Card:

  • Press ✎ Correct under the Card. It shows as a tag above the input field, and you write what to change, such as “Show the regions as horizontal bars”.
  • Or type @ in the input field. The list shows the Cards of the Conversation first, above your files and folders; type a few letters of a title to narrow them down. Choose a Card with the arrow keys and press Enter, or click it: it is tagged above the input field and leaves nothing in your text. Escape closes the list.

You can tag several Cards in one message, and remove a tag with its ×. The tags go with your next message: the agent gets each tagged Card as you see it now, with the options you set, and answers with a new Card. The Conversation keeps both, so you can compare them.

A Card doesn’t have to come from a query. The agent also shows as a Card:

  • a file its code wrote in the Sandbox, such as a CSV file of figures it worked out. The Card’s header says From the file and the file’s name, and the Source tab says that the agent’s code wrote the rows.
  • what an Integration answered, when the answer is a list of rows. The Card’s header names the Integration, and the Source tab names the tool the agent called.

Such a Card can’t go on a Dashboard: a Dashboard refreshes its Cards by querying their Data Source again, and these rows came from somewhere else. Its + Dashboard button is disabled, and the Card says to load the data into a Managed Database first.

+ Dashboard below a Card puts a copy of it on a new Native Dashboard or on one of yours, where it queries its Data Source again whenever you refresh it. See Dashboards.

The agent never writes numbers into a chart. The chart reads them from the result, so what you see is what the database, the code or the Integration returned, not what the model remembered. A chart whose spec holds numbers of its own is refused, and the agent answers in words instead.

A Card keeps at most 10,000 rows, and its chart draws at most 5,000 points, so it stays fast in your browser. If a result is larger, the Card says so instead of a chart, and the agent adds up its data in the query and tries again.

A Card keeps the rows it showed. If you can no longer query its Data Source, or the Data Source was deleted, the Card still shows its chart, table and query, and says that it shows the rows as they were when it was made.

Cards stay as long as their Conversation. When you delete the Conversation, its Cards go with it for good.

Below the chart, a screen reader reads what the chart shows in words: its type, how many rows it has, what each axis shows and, where it has one measure, its highest and lowest value. The Table tab has every row.